People who look alike also have similar DNA. This has been scientifically proven.

James from Aofei Temple
Quantum Bit | Public Account QbitAI

"Hey, you look a lot like my ex-girlfriend, I almost made a mistake."

When boys talk to girls, there are always people who use these cliché tricks to make everyone seem familiar and destined (manual dog head).

Of course, it's not just about real-life chatting. The fact that they look like "half-brothers and sisters" has always been a hot topic for online meme...

For example, Lin Donglin and Guo Junjie...

Not only you and I are concerned, this matter has been used by scientists as a research problem, and they have come to a conclusion:

People who look similar are indeed "connected"; they may have similar DNA.

A recent article published in a Cell journal shows that people with similar faces share the same gene sequences.

How did they come to this conclusion? Is it reliable?

Lets come look.

Ask the machine to find the "half-brother"

First, we need to find a bunch of people who look almost identical as samples. Faced with a world population of 7 billion, how to find them is a problem.

The scientists found Canadian artist François Brunelle, who has been collecting similar faces from around the world since 1999.

Relying on "finding the right way", scientists obtained 32 pairs of similar-looking samples and asked participants to provide detailed biometrics, location and lifestyle questionnaires.

In order to ensure a more objective face comparison, scientists found three algorithms/models to jointly "disassemble" facial details and determine similarity.

These include -

(1) Custom deep convolutional neural network Custom -Net , a CNN model from the AI ​​company Herta, mainly used in the monitoring field;

(2) MatConvNet , which is commonly used in Matlab for face classification tasks ;

(3) Microsoft Oxford Project Face API from Microsoft Azure , commonly used for facial analysis;

These models have millions of built-in parameters, and have been trained (folded) (honed) on millions of facial images under thousands of themes. They can identify postures, hairstyles, expressions, age, accessories and many other features. For similar images, the model A similarity interval from 0 to 1 will be given.

If the machines "jointly judge" that the samples are extremely similar, the scientists will focus on the samples. In the end, 16 of the 32 pairs of samples were recognized by the three parties.

Next comes the DNA comparison phase.

The scientists used a genetic signature called single nucleotide polymorphism (SNP) to compare DNA samples from the saliva of groups of people who looked very similar.

After that, unsupervised clustering heat map detection is used to let the machine determine whether each pair of DNA belongs to the same type (i.e., similar).

The conclusion is a bit due to Sting

The results showed that among the 16 pairs of sample DNA "all recognized by the machine", scientists found that 9 pairs were clustered into one category. They shared more than 19,200 SNPs in 3,700 genes. Similarly, in the K-means algorithm, the final conclusion is similar.

In addition, scientists also studied the biological characteristics and lifestyles of similar pairs. They were also correlated with each other in the paired groups in terms of weight, height, smoking, education level, etc., which shows that shared genetic variations not only affect appearance, but also cause similar behaviors and habits.

The researchers described the results as "striking". The researchers also ruled out the possibility of close relatives and found that the samples were more genetically similar than identical twins.

What is more interesting is that among the remaining 16 groups of samples that failed to pass the third-party machine vision feature recognition, only one pair of sample DNA was clustered in one set.

In terms of race and region, scientists also conducted ancestral tracing. They observed that the ancestral areas of almost all similar paired samples were very close to each other.

However, it should be pointed out that the topic of this article has different conclusions in other multi-omics studies.

For example, in the DNA methylation model, only one pair of 16 pairs of highly similar samples matched;

From a microbial perspective, among the 16 pairs of similar samples, only one pair had similar oral flora, and this pair of samples was not clustered together through SNP. Some studies in this field believe that the similarity of oral microorganisms is related to subcutaneous fat, which may lead to people with similar appearance (such as a fleshy face) to have similar oral microbial characteristics.

team introduction

The first author of this article, Ricky S. Joshi, is a computational biologist from the Josep Carreras Leukemia Institute (IJC) in Barcelona, ​​Spain. His research focuses on genetics and epigenetics. His personal page shows that Ricky is committed to discovering patterns in the human genome. Pathological variation.

The second author, Maria Rigau, is also based in Spain. She is a PhD student at the Barcelona Supercomputing Center (BSC) majoring in life sciences. Her homepage avatar shows that she is a female researcher who quite likes dogs.

One More Thing

Some netizens pointed out that Musk looks a bit like Edison, and posted the following picture :

What do you think? (dog head)

Reference links:
https://www.cell.com/cell-reports/fulltext/S2211-1247(22)01075-0
https://neurosciencenews.com/genetics-look-alikes-21283/
https://hertasecu rity .com/
https://bigthink.com/health/look-alike-genetics-behavior/

click here